Question about attacking to the ResNet

[yxwang-10]

Hi,
I try to perform the Autozoom-ae attack on the ResNet with was pretrained by the tensorflow .But i have some problems，i modified the prediction function so that the value of the images between 0 and 255 and also reduce the men of the imagenet dataset,but when run the iteration of the attack,the loss1 is always 0.